Hmga2 regulates self-renewal of retinal progenitors

The vertebrate retina is a simple and accessible central nervous system (CNS) model, facilitating our understanding of how the brain develops and functions. It consists of seven different cell types generated by an evolutionarily conserved, temporal sequence from single multipotential progenitors (Robinson, 1991; Livesey and Cepko, 2001). For example, barring certain overlaps, in most species the retinal ganglion cells (RGCs), cone photoreceptor (CPs), horizontal cells (HCs) and the majority of amacrine cells (ACs) are born during the early stage of histogenesis, whereas the rod photoreceptors (RPs), bipolar cells (BCs) and Müller glia (MG) are generated during the late stage of histogenesis (Rapaport et al., 2004). A variety of approaches in different species have revealed the identity of cell-intrinsic (Cepko, 1999; Hatakeyama et al., 2001; Agathocleous and Harris, 2009) and cell-extrinsic (Cepko, 1999; Yang, 2004; Agathocleous and Harris, 2009) factors in the generation of specific retinal cell types. A general theme has emerged that contextual interactions between these factors regulate when, and along which sublineage RPCs would differentiate.

In humans, retinal cells are born over 28 weeks of gestation; in rats, the animal studied here, retinal histogenesis takes half of the gestation period plus two postnatal weeks to complete (Robinson, 1991). Such a prolonged period of histogenesis likely requires the progenitors to self-renew in order to sustain stage-specific generation of different cell types. Although the multipotential nature of the RPCs has been extensively examined (Livesey and Cepko, 2001), little is known about their self-renewal property. In vitro studies examining the differentiation potential of RPCs from late-stage rat retinal histogenesis using time-lapse microscopy, and retrospective morphological and immunocytochemical identification demonstrated asymmetrical divisions that generated differentiated and undifferentiated cells (Cayouette and Raff, 2003; Gomes et al., 2011). Since asymmetrical division is how stem cells maintain their population, this could be evidence of the self-renewal ability of RPCs. However, stem cells that have advanced along the differentiation path and transitioned into progenitors/precursors may also divide asymmetrically, thereby blurring the boundary between them and stem cells if features such as stem cell-specific markers and propagation of clones over generations are not considered. When late RPCs are examined on these criteria in a neurosphere assay, which is used to test neural stem cell (NSC) self-renewal elsewhere in the CNS (Weiss et al., 1996; Temple, 2001), they fail to generate colonies at lower density, precluding tests to determine their ability to generate successive colonies and if they retain the multipotentiality of their parents (Ahmad et al., 2004). A conservative interpretation is that the self-renewal property of RPCs is not entirely intrinsic, and standard mitogens such as epidermal growth factors (EGF) and fibroblast growth factors 2 (FGF2) used in the assay are inadequate in exposing this property in vitro. This interpretation confers an important role on the environment in regulating the RPC self-renewal property.

Here, we test the hypothesis that the self-renewal property of RPCs is non-cell autonomous, requiring contributions from other cells in the histogenic environment. Without the identity of contributing cells in the developing retina, this premise was examined in the presence of the endothelial cells (ECs), which are known to support stem cells in neural (Palmer et al., 2000; Shen et al., 2004,, 2008; Imura et al., 2008) and extra-neural (Yin and Li, 2006; Butler et al., 2010) niches. As evidence suggests different stem cells use conserved mechanisms for their maintenance (Morrison and Spradling, 2008; Shenghui et al., 2009), we argued that ECs may sustain RPC self-renewal in vitro, thus revealing the regulatory mechanism(s) underlying their maintenance. We observed that, in the presence of ECs, RPCs generated successive neurospheres at low density and retained parental multipotentiality, thus fulfilling the self-renewal criterion. ECs similarly influenced RPCs in vivo, demonstrated by higher indices of cell proliferation, neurosphere generation and a side population (SP)-cell phenotype upon intravitreal injection of EC-conditioned medium (ECCM). Furthermore, we demonstrate that Hmga2 (Noro et al., 2003), which is expressed in the developing retina, intrinsically mediated the environmental influence on RPC self-renewal. Thus, our results offer a framework to integrate diverse intercellular influences by epigenetic regulators in the dynamic stem cell niche of the developing retina. Additionally, these results may help formulate approaches to overcome inefficient expansion of RPCs in vitro, a significant barrier to cell therapy.

Although embryonic RPCs reside in an avascular developing retina, we asked whether their self-renewing property could be revealed through their interactions with ECs (Shen et al., 2004). We cultured embryonic day 18 (E18) retinal cells, representing the beginning of late-stage histogenesis, at different densities in the presence of ECCM or EGF (control) for 5 days and examined the generation of neurospheres. In low-density culture (≤3×10⁴ cells/cm²) neurospheres were detected only in the presence of ECCM (Fig. 1A-C). Subsequent cell culture at higher densities caused neurosphere generation in both conditions; however, neurospheres generated with ECCM were significantly higher in number and size compared with those without ECCM (Fig. 1A-C). The terminal deoxynucleotidyl transferase dUTP nick-end labeling (TUNEL) assay revealed no significant difference in cell survival between the two cultures conditions, thus eliminating selective ECCM-mediated survival as the cause (Fig. 1D-F). We then asked if the difference in the numbers and size of neurospheres were reflected in cell proliferation and progenitor properties. First, RPCs were cultured with the nucleotide analogue 5-bromo-2-deoxy uridine (BrdU) at a plating density (1.0×10⁵ cells/cm²) that allowed the generation neurospheres in the control groups for comparison. Results showed a significant increase in the BrdU⁺ cell number when cultured in ECCM (40.4±0.1) versus controls (25.2+0.2; P<0.05) (Fig. 1G). Second, examining the expression of select cell-cycle regulators revealed increased Ki67 (Mki67), Pcna and cyclin D1 (Ccnd1) transcript levels in ECCM neurospheres versus controls (Fig. 1H-J). Last, a concomitant increase occurred in the number of BrdU⁺ cells expressing three different retinal progenitor markers (Pax6, Rx and Chx10) in ECCM neurospheres versus controls (Fig. 1K-Q). Together, these observations suggested that ECCM contained activities that supported the amplification of RPCs. Whether or not the proliferating RPCs could self-renew was examined using three different approaches. First, we subjected RPCs to ex vivo assessment of the self-renewal of neural stem cells (Ferron et al., 2007). Cell dissociated from E18 retina were cultured in the presence and absence of ECCM at densities of 5, 10 and 20 cells/μl. Primary neurospheres were obtained at 20 cells/μl, and only with ECCM. Generation of secondary and tertiary neurospheres was obtained at 5 cells/μl with ECCM, and demonstrated a progressive increase in their numbers (Fig. 2A). Second, cells in both primary and secondary neurospheres were subjected to identical differentiation conditions, where they differentiated into generic neurons and glia in a similar proportion. This suggested the retention of parental multipotentiality over successive generations (Fig. 2B). To determine whether EC exposure altered the ability of RPCs to generate late-born retinal neurons, control and ECCM-generated neurospheres were cultured with postnatal day1 (PN1) retinal cell conditioned medium to promote differentiation along late retinal sublineages (Parameswaran et al., 2010). No significant difference in rod photoreceptor (e.g. rhodopsin)- and bipolar cell (e.g. mGluR6)-specific transcripts levels was observed, suggesting the intrinsic differentiation property of progenitors remained unchanged (Fig. 2C,D). Third, to determine the frequency of self-renewing RPCs and cellular interactions needed for their maintenance during late histogenesis, limiting dilution analysis (LDA) was performed, which revealed a 0.078% frequency of such cells (Fig. 2E). Amplification in cell numbers is necessary to generate clonal neurospheres; the increase in successive neurospheres points to the adoption of symmetrical cell division by self-renewing RPCs. To test this premise, we examined temporal generation of SP cells, a functional characteristic of stem cells and also of RPCs (Bhattacharya et al., 2003), in neurosphere assays. We observed that SP cell numbers increased ∼7-fold at day 5 and ∼3-fold at day 8, compared with day 2 and day 5, respectively (Fig. 2F-H), a temporal increase in number likely sustained by symmetrical cell division. Together, these observations suggested that ECCM activities allowed for the successive generation of RPCs at low density through symmetrical cell division, while retaining parental properties of multi-lineage differentiation, i.e. self-renewal.

Next, to determine whether the influence of ECs on self-renewal was not simply a function of the neurosphere assay, the effects of ECCM were examined on RPCs in vivo. Concentrated ECCM was injected intravitreally into the eyes of PN1 pups. RPCs were examined for proliferation and progenitor properties two days later, at PN3 (Fig. 3A). The following two controls were used: (1) vehicle controls in which pups were injected retinal culture medium (RCM); and (2) activities controls in which ECCM pre-incubated with antibodies against FGF2, stem cell factor (SCF) and pigment epithelium-derived factor (PEDF) to neutralize the specific signaling pathways (i.e. nECCM). The rationale for neutralizing these pathways was based on microarray analysis (see below) and availability of neutralizing antibodies. Prospective enrichment of RPCs as SP cells from the injected retina demonstrated an ∼9-fold increase in the proportion of retinal SP cells in ECCM-injected eyes, compared with RCM controls (0.006% versus 0.052%) (Fig. 3B,C). The positive effect of ECCM on retinal SP cells was abrogated in eyes injected with nECCM (0.052% versus 0.008%) (Fig. 3C,D). Similarly, the number of proliferating cells (BrdU⁺ cells), which showed a significant increase in ECCM-injected eyes compared with RCM controls (25.7±0.05 versus 20.7±0.15, P<0.05), decreased significantly in the nECCM-injected eyes (21.2±0.05 versus 25.7±0.05, P<0.05) (Fig. 3E). Next, we determined whether the influence of ECCM on RPCs in vivo was sufficient to allow them to generate neurospheres with EGF, a non-conducive condition for generating neurospheres at low density. To account for the expected exhaustion of RPCs at this late stage of histogenesis, the plating density of PN3 cell dissociates was increased. We observed an ∼3-fold increase in the number of neurospheres in ECCM-exposed cell dissociates versus RCM controls (96.0±10.6 versus 34.2±15.6, P<0.05). The effect of ECCM on neurosphere generation was abrogated when ECCM was neutralized prior to injection (44.7±7.9 versus 96.0±10.6, P<0.05) (Fig. 3F). Together, these results suggested that ECCM activities had regulatory effects on RPCs in vitro and in vivo.

To understand the mechanism underlying EC-mediated regulation of RPCs, we performed hypothesis-driven transcriptional profiling of ECCM and control neurospheres. Our hypothesis was, given the cell-extrinsic influence, that the mechanism for self-renewal involves the expression of genes encoding intercellular signaling components and universal regulators of genes for nuclear integration of cell-cell signaling. Under this premise, a general increase and decrease in facilitators of the cell cycle and differentiation, respectively, would determine outcomes of regulation under the influence of ECs. Microarray analysis revealed that the exposure of the neurospheres to ECCM suppressed genes that encode regulators (e.g. Crx, Nrl and Neurod1) and markers [rhodopsin, recoverin, mGluR6 (Grm6) and Rlbp1] of different retinal cell types. By contrast, ECCM exposure resulted in upregulation of genes that regulate the cell cycle (e.g. Ccnd1 and ccna1), cell commitment (e.g. Id1, Id2, Id4, Hes1 and Lef1) and epigenetic changes for the self-renewal (e.g. Hmga2 and Smarca4). From the perspective of the cell-cell signaling, a noteworthy increase occurred in the expression of receptors for Notch (Notch1), FGF (Fgfr1), VEGF (Vegfr2), Wnt (Fzd5), SCF (Kit), BMP (Bmpr2) and activin (Acvr1) pathways with ECCM (Fig. 4A). The differential expression patterns of genes, confirmed by Q-PCR analysis for select genes (supplementary material Fig. S1), suggested that ECCM activities regulated RPCs by recruiting disparate signaling pathways whose targets could be Hmga2 and Smarca4. Here, given its role in regulating tissue-specific stem cells, we examined the involvement of Hmga2 (Nishino et al., 2008; Copley et al., 2013). Hypothesis testing involved: (1) the examination of intercellular pathways, which necessitates that ECs express the corresponding ligands for the identified receptors in RPCs (Wang et al., 2007); and (2) determining whether the recruitment of the signaling pathway(s) engaged Hmga2 expression. Examination of transcripts by reverse transcriptase polymerase chain reaction (RT-PCR) revealed expression of genes Kit, Fgfr1, Lrp6, Vegfr1, Vegfr2, Bmpr1, Bmpr2, Actr2a, Actr2b and Notch1, and Frizzled genes in E18 RPCs, and transcripts corresponding to their ligands (Scf, Fgf2, Wnt2b, Vegfa, Bmp4, activin and follistatin) in ECs (Fig. 4B, supplementary material Fig. S2). PEDF was included in the analysis given its own expression and that of its receptor (PEDFR), in the ECs and RPCs, respectively (Fig. 4B), and the known influence of PEDF on proliferation of cortical progenitors (Ramirez-Castillejo et al., 2006; Andreu-Agullo et al., 2009) and RPCs (De Marzo et al., 2010). To test the involvement of the signaling pathways, we studied the generation of secondary neurospheres in the presence of their inhibitors. Results showed that attenuating each of the signaling pathways significantly reduced neurosphere numbers but did not eliminate their generation, suggesting each pathway was important and might act in concert to regulate self-renewal (Fig. 4C). To test this premise, cells were cultured in the combined presence of inhibitors (cocktail of inhibitors) of these pathways. Neurospheres were generated, albeit at reduced numbers, confirming the interactions between these pathways, but also revealing the possible involvement of other unknown pathway(s). TUNEL staining revealed that the difference in neurosphere generation was not due to selective survival under control conditions (supplementary material Fig. S3). That these pathways engaged Hmga2 was demonstrated as follows. First, we examined the expression of Hmga2 during neurosphere formation in the presence of ECCM and ECCM+cocktail inhibitors. Robust expression levels of Hmga2 transcripts were observed in ECCM-generated neurospheres, which were abrogated with cocktail inhibitors (Fig. 4D). Second, we determined levels of Hmga2 expression in vivo following intravitreal injection of ECCM/nECCM and RCM as described above. We observed ∼3-fold increase in the proportion of cells expressing Hmga2 immunoreactivities in ECCM-treated retina versus RCM controls (45.8±4.36 versus 17.42±4.40, P<0.05). No significant change occurred in the proportion of Hmga2⁺ cells between nECCM- and ECCM-treated retina (32.3±6.20 versus 45.8±4.36, P>0.05) (Fig. 4E). Together, these results suggested that the ECCM-mediated activation of signaling pathways regulated stem cell properties of RPCs in concert with Hmga2 in vitro and in vivo.

Hmga2 involvement in the regulation of stem cell properties of RPCs necessitates the association of Hmga2 expression with the temporal aspects of retinal histogenesis. Therefore, we first determined temporal and spatial expression patterns during retinal development. Hmga2 transcripts were observed at early histogenesis at E14, and transcript levels declined significantly at the beginning of late histogenesis (E18) (Fig. 5A). At PN3, Hmga2 transcript levels were lowest compared with earlier stages and not detectable in adult retina. Spatial localization of Hmga2 immunoreactivities in E14 and E18 retina revealed their distribution in the outer neuroblastic layers, predominantly in BrdU⁺ cells, with a peripheral-to-central gradient (Fig. 5B,C). The temporal and spatial expression patterns suggested that Hmga2 was physiologically associated with RPCs and that the decline in Hmga2 expression corresponded to progressive exhaustion of RPCs, as differentiation completed. To further establish an association of Hmga2 with RPC self-renewal, we examined expression of Hmga2 and its negative regulator Let7 (Lee and Dutta, 2007) in ECCM-mediated neurosphere assay. First, we observed a significant increase in Hmga2 transcript and protein levels (Fig. 5D-G) and decrease in Let7 miRNA levels (Fig. 5H). Second, levels of Let7 miRNA increased and that of Hmga2 transcripts decreased precipitously when cocktail inhibitors compromised neurosphere generation (Fig. 5I,J), suggesting Hmga2 mediated the influence of ECCM on the RPCs. Last, we examined the expression of Junb and P19^arf, components of the emerging Hmga2 axis for regulating self-renewal of NSCs (Nishino et al., 2008). Junb and P19^arf transcript levels increased significantly with a cocktail of inhibitors versus controls. This was in accordance with the previous observations that Hmga2 inhibits Junb expression to negatively regulate P19^arf (Nishino et al., 2008) (Fig. 5K,L). Together, these observations suggested that the Hmga2 regulatory axis was active in RPCs in development and associated with ECCM-mediated self-renewal.

Next, we perturbed Hmga2 expression in vitro and in vivo to determine Hmga2 involvement in EC-mediated RPC regulation (Fig. 6A). First, the gain-of-function experiments (validated in supplementary material Fig. S6A-D) were carried out in vitro, where E18 retinal cells were transduced with a dual-promoter lentivirus that expressed Hmga2+GFP or only GFP (minus Hmga2) as a control (Nishino et al., 2008) (see Materials and Methods; supplementary material Table S3), and subjected to neurosphere assay. GFP⁺ cells, which co-express Ki67/Pax6, in cell dissociates represented transduced RPCs (Fig. 6B; supplementary material Fig. S7). In addition, we observed a significant increase in the numbers of neurospheres in gain-of-function groups versus controls (Fig. 6C); the gain-of-function neurospheres also contained significantly more GFP⁺ Ki67⁺ (Fig. 6D)/GFP⁺ Pax6⁺ (Fig. 6E) cells than controls, demonstrating a positive influence of Hmga2 overexpression on RPC regulation, which was independently corroborated by a significant increase in BrdU⁺ (Fig. 6F) and SP (Fig. 6G) cell numbers in gain-of-function neurospheres versus controls. Additionally, changes in response to Hmga2 overexpression were accompanied by a significant decrease in levels of Junb/p19^arf transcripts (supplementary material Fig. S4A-D), suggesting involvement of the Hmga2-Junb/p19^arf axis in RPC regulation. Next, to corroborate Hmga2-mediated regulation of RPCs in vivo, Hmga2 (3′UTR del)+GFP/control-GFP lentivirus was injected intravitreally into PN1 pups. Because the expression of Let7b, the negative regulator of Hmga2 transcripts, inversely correlates with the decline of Hmga2 expression postnatally, 3′UTR deleted Hmga2 lentivirus constructs were used to stabilize Hmga2 against Let7b (Nishino et al., 2008). Retinae were recovered at PN4 and dissociated to examine RPCs properties. We observed a significant increase in the numbers of GFP⁺ Ki67⁺ (Fig. 6H)/GFP⁺ Pax6⁺ (Fig. 6I)/BrdU⁺ (Fig. 6J) cells in gain-of-function groups versus controls, suggesting that RPCs were similarly influenced by Hmga2 overexpression in vivo. In addition, RPCs from gain-of-function retina generated ∼1.5 fold more neurospheres (Fig. 6K), which contained significantly more SP cells (Fig. 6L) than did controls.

When retina were examined at PN10 for differentiation by immunofluorescence analysis, no difference was observed in rod photoreceptor generation (Fig. 6M,N) or other late-born cells, i.e. the bipolar cells and Müller glia (supplementary material Fig. S4). To determine whether the lack of effect on differentiation was due to perturbing Hmga2 expression at a stage (PN1), where most RPCs were committed, or to low efficiency of transduction, or to both, we performed gain-of-function experiments in E18 retinal explants; the effects on RPC regulation were examined after 4 and 10 days in culture. As observed in vitro and in vivo, numbers of GFP⁺Ki67⁺ (Fig. 6O)/GFP⁺Pax6⁺ (Fig. 6P)/BrdU⁺ (Fig. 6Q)/SP(R) cells increased at day 4 in gain-of-function groups versus controls. When explants were examined at day 10 for differentiation, in contrast to in vivo experiments, we observed a significant decrease in the numbers of rod photoreceptors in gain-of-function groups versus controls, suggesting that sustained Hmga2 expression negatively influences differentiation along the rod photoreceptor lineage (Fig. 6S-U). No such differences were observed between the groups in bipolar cells (PKC⁺ cells) and Müller glia (GLAST⁺ cells) generation (supplementary material Fig. S4). Next, we carried out the loss-of-function experiments, which were similar to gain-of-function experiments (Fig. 6A,B; supplementary material Fig. S7), except for viral transduction that included a dual-promoter Hmga2 siRNA+GFP/control GFP lentivirus (see Materials and Methods and supplementary material Table S3) to attenuate Hmga2 expression (validated in supplementary material Fig. S6E-G). When transduced E18 retinal cells were subjected to a neurosphere assay in the presence of ECCM, unlike gain-of-function results, significantly fewer neurospheres were generated in the loss-of-function group versus controls (Fig. 7A). In addition, the loss-of-function neurospheres contained significantly fewer Ki67⁺GFP⁺ (Fig. 7B)/Pax6⁺GFP⁺ (Fig. 7C)/BrdU⁺ (Fig. 7D)/SP (Fig. 7E) cells compared with controls. However, levels of Junb/p19^arf transcripts were higher in the former versus the latter, suggesting the involvement of the Hmga2-Junb/p19 axis in RPC regulation (supplementary material Fig. S5). When cell dissociates from in vivo-transduced retina were examined, a significant decrease in GFP⁺ Ki67⁺ (Fig. 7F)/GFP⁺ Pax6⁺ (Fig. 7G)/BrdU⁺ (Fig. 7H) cell numbers was observed in loss-of-function groups versus controls, suggesting that RPC were similarly influenced by the attenuation of Hmga2 expression in vivo. As observed in vitro, RPCs from loss-of-function groups generated significantly fewer neurospheres (Fig. 7I) that also contained fewer SP cells (Fig. 7J) than controls. However, as in the gain-of-function approach, no significant differences were observed in the generation of late-born cells in vivo (Fig. 7K,L; supplementary material Fig. S5). Based on the rationale above, when the loss-of-function experiments were performed on E18 retinal explants, there was a significant decrease in the numbers of GFP⁺Ki67⁺ (Fig. 7M)/GFP⁺Pax6⁺ (Fig. 7N)/BrdU⁺ (Fig. 7O)/SP (Fig. 7P) cells at day 4 in culture in loss-of-function explants versus controls, as observed both in vitro and in vivo. However, in contrast to in vivo experiments, when explants were examined at day 10 in culture, there was a significant increase in the numbers of rod photoreceptors in loss-of-function groups versus controls (Fig. 7Q-S). No significant differences in the number of bipolar cells (PKC⁺ cells) and Müller glia (GLAST⁺ cells) were observed between the two groups (supplementary material Fig. S5). Together, these results suggest that Hmga2 regulates RPCs and is functionally involved in mediating cell-extrinsic influences on their self-renewal.

RPCs, whether from early or late stages of retinal histogenesis, tend not to generate typical neurospheres or clones in low-density cultures, but instead produce floating cell aggregates (Ahmad et al., 2004) or fewer than a dozen adherent clones (Jensen and Raff, 1997). Neither aggregates or clones generate secondary clones, failing the self-renewal test. The apparent lack of self-renewal in RPCs in vitro, compared with their counterparts from other CNS regions (Weiss et al., 1996; Temple, 2001), is counterintuitive to their maintenance being required to sustain prolonged histogenesis, a significant proportion of which is completed postnatally. Here, we demonstrate that RPCs self-renew, but this property probably depends on the microenvironment. For example, under the influence of ECs, which are known to support stem cells in a variety of niches (Palmer et al., 2000; Shen et al., 2004,, 2008; Yin and Li, 2006; Imura et al., 2008; Butler et al., 2010), RPCs generate clonal neurospheres. These neurospheres can be passaged; cells therein possess the multi-potentiality of the parents, thus fulfilling self-renewal criterion in vitro. The technical limitations of serial transplantation of neural progenitors preclude the unambiguous determination of RPCs self-renewal in vivo. However, the increased number of proliferating progenitors and their ability to form neurospheres under non-conducive conditions in response to ECCM, and the abrogation of these effects upon enforced attenuation of Hmga2 expression, attest to their non-cell autonomous self-renewal in vivo and identifies some of the mechanisms involved. The frequency of such progenitors in E18 retinal cell dissociates is 0.078%. The low frequency of self-renewing RPCs might reflect their exhaustion by E18, when the majority of cells are committing along the predominant rod photoreceptor lineage.

An interesting question is how the diffusible factors recruit cell-intrinsic machinery to promote cell proliferation and maintain multipotentiality. In addition, what is the source(s) of these factors in the developing retina? There are several candidate sources for these factors. For example, within the retina, factors expressed by cells in different stages of differentiation may influence RPC maintenance; Shh expressed by nascent RPCs promotes RPC proliferation. This is presumably to maintain the RPCs for subsequent rounds of differentiation (Wang et al., 2005). Without the retina, retinal pigment epithelium (RPE) may play a significant role in regulating self-renewal, as it is a source of Wnt2b (Cho and Cepko, 2006), Vegf (Sonoda et al., 2009) and Pedf (Serpinf1) (Tombran-Tink et al., 1991; Sonoda et al., 2009). The lens, the development of which precedes formation of the optic cup and is regulated by FGF signaling, may be a source of Fgf2 for retinal development. Last, although the developing retina is avascular, the choroidal circulation, which is established during formation of the optic cup (Saint-Geniez and D'Amore, 2004), may serve as a source of regulatory factors. However, regardless of sources, these factors may not individually regulate RPC self-renewal. It is likely that their combined effects with unknown factor(s) alter the expression of intrinsic factors, such as Hmga2, to specific levels required for self-renewal. Recently, Hmga2 was shown to repress Junb and indirectly reduce expressions of p16^ink4a and p19^arf, thereby increasing self-renewal of multi-lineage NSCs (Nishino et al., 2008). The role of Hmga2, however, was limited to young mice because Let7 expression increases with age, leading to a reciprocal decline in Hmga2 expression and consequently self-renewal (Nishino et al., 2008). Several observations support that the Hmga2 regulatory axis is involved in a similar capacity in RPCs. First, the developmental decline in Hmga2 expression (localized predominantly in proliferating RPCs) corresponds to progressive exhaustion of RPCs, implicating Hmga2 in RPC maintenance. Second, changes in Hmga2 expression in early postnatal retina following exposure to ECCM/nECCM are associated with changes in the proliferation index and retinal SP cell number. Third, examining RPCs at late-stage histogenesis revealed that the unmasking of their self-renewal property by ECs accompanied reciprocal expression patterns of Let7, Hmga2, Junb and p19^arf, as expected for the Hmga2-regulatory axis in self-renewal. Finally, attenuation of Hmga2 expression compromised EC-dependent RPC self-renewal, whereas ectopic expression of Hmga2 increased the indices of self-renewal, in vitro and in vivo. This functionally implicates Hmga2 as an intrinsic factor regulating non-cell autonomous RPC self-renewal.

However, the observation that RPC self-renewal depends upon modulation of Hmga2 transcript levels, not their presence alone, points to the significance of stoichiometry in Hmga2 expression levels for cells acquiring the self-renewal capacity. This mechanism appears similar to the interactive nexus genes, Oct4 and Nanog, in the pluripotency network, the expression levels of which must reach a specific threshold before pluripotency is realized (Pan et al., 2006). Thus, a similar network can be proposed for RPC self-renewal with interactive nexus genes, one of which is likely to be Hmga2. Other genes occupying the nodes in this network may be the chromatin remodeling ATPase Brg1 (also known as Smarca4) and Bmi1, a polycomb gene. Both Smarca4 and Bmi1 regulate stem-cell self-renewal (Kidder et al., 2009; Chatoo et al., 2010), are expressed in the developing retina and are modulated in RPCs in response to EC exposure (S.P., X.X. and I.A. unpublished). These epigenetic factors likely integrate the influence of the niche, rendered through disparate signaling pathways, and globally alter transcription of downstream genes required for self-renewal. Thus, our results provide a framework for the integration of diverse intercellular influences by epigenetic regulators in a dynamic stem cell niche in the developing retina. Furthermore, these findings allow for the formulation of approaches to maintain RPCs in sufficient numbers in vitro, which is essential for the practical use of RPCs for therapeutic purposes.

The Institutional Animal Care and Use Committee (IACUC) at University of Nebraska Medical Center (UNMC) (protocols #97-100-08FC and #95-005-09FC) approved the study. Animals were housed in the Department of Comparative Medicine at UNMC. All experiments were carried out on timed pregnant Sprague Dawley rats obtained from SASCO.

The neurosphere assay was as described previously (Ahmad et al., 1999; Parameswaran et al., 2010). Briefly, cell dissociates from E18 retina were cultured in RCM supplemented either with EGF (20 ng/ml)/EGF+ECCM for 4-5 days to generate neurospheres. Retinal differentiation was induced with PN1 CM. For the clonal density assay (Ferron et al., 2007), cells were plated in control or ECCM conditions at 5,10 and 20 cells/µl. Primary and secondary neurospheres were manually titrated and cultured at 5 cells/µl under above-mentioned conditions to obtain secondary and tertiary neurospheres, respectively. PCR and immunofluorescence analyses were performed using gene-specific primers (supplementary material Table S1) and primary antibodies (supplementary material Table S2), respectively. Measurements were performed in triplicates. Western analysis performed as previously described (Das et al., 2007).

Retinal explant cultures were performed as previously described (Del Debbio et al., 2010). Briefly, E18 retinae were placed with the ganglion cell layer (GCL) side upwards on a 0.4 µm semi-permeable membrane (Millipore) and cultured in RCM with 5% FBS. Explants were exposed to BrdU (10 μM) for 8 h on day 4 and collected for analyses at the end of day 4 and day 10. Sections of explants and that of age-matched retina (PN5; taking into consideration explants from E18 retina, birth at the gestation age 22 and the duration of the culture) were examined for the laminar structure to determine the integrity of the explant culture (supplementary material Fig. S8).

The details of the lentiviral vectors used in the study are provided in supplementary material Table S3. The backbone of Hmga2+GFP and Hmga2 (3′UTR DEL)+GFP and control GFP viral vectors was the dual promoter viral plasmid, pLentilox RSV, in which Hmga2 and GFP expression was driven by RSV and CMV promoters, respectively (Nishino et al., 2008). The backbone of Hmga2 siRNA GFP and control (scrambled) siRNA GFP vectors was the viral plasmid piLenti siRNA GFP (ABM), where the opposing polymerase III promoters (H1 and U6) transcribed the sense and antisense strands of the siRNA duplex and a CMV promoter driving GFP expression. The target sequence for rat Hmga2 siRNA was 5′AGACCCAAAGGCAGCAAGAACAAGAGCCC3′. The sequence for the scrambled siRNA control was 5′GGGTGAACTCACGTCAGAA3′. In the mCherry control vector (GeneCopoeia), GFP expression was driven by the CMV promoter.

Lentivirus preparation and transduction were as previously described (Parameswaran et al., 2012). Given the RPC limitations, the efficacy of Hmga2 overexpression and siRNA-mediated gene silencing was determined in 293T and C6 glioma cells, respectively. Lentivirus-transduced neurospheres/retina/retinal explants were dissociated and cells plated on glass coverslip and subjected immunofluorescence analysis to quantify transduced RPCs. The perturbation experiments were carried out three times in triplicates as follows: four animals/group (in vivo perturbations), nine retinae/group (ex vivo perturbations) and 10-12 E18 embryos/group (in vitro perturbations).

Injections were performed on PN1 pups as described previously (Das et al., 2006). RCM and ECCM media were concentrated using 10 kDa Amicon filters. Concentrated ECCM, neutralized with anti-FGF (2.5 μg/ml), anti-PEDF (2.5 μg/ml) and anti-SCF (1 μg/ml) antibodies for 4 h at 37°C was injected in a 1 µl volume. For Hmga2 gain-of-function studies, 1 µl of Hmga2 (3′UTR DEL)+GFP/Control GFP lentivirus (1×10¹⁰ IFU/ml) were injected. For Hmga2 loss-of-function studies, 1 µl of Hmga2 siRNA-GFP/control GFP lentivirus (1×10¹⁰ IFU/ml) mixed 1:1 with concentrated ECCM were injected.

Bovine pulmonary artery EC lines (BPAE) (ATCC) were maintained in Eagle's minimum essential medium (EMEM) with 20% FBS. Cells were plated at the density of 5×10⁴ cells/cm² in RCM. After 3 days, CM was collected, centrifuged to remove floating cells and filtered and stored at −80°C until use.

LDA analysis was performed as described previously (Das et al., 2006). Briefly, retinal cells in serial dilutions in 200 μl aliquots were plated individually in a 96-well plate and cultured for 7 days. Fraction of wells, lacking neurospheres were scored. The negative logarithm of these wells was plotted against the number of cells/well. The zero term of the Poisson equation predicted that when 37% of test wells were negative, one stem cell/well existed (on average).

To analyze signaling pathway involvement, primary neurospheres were dissociated and cultured (4.5×10⁴ cells/cm²) in ECCM containing individual inhibitors [DAPT (5 mM/ml), DKK1 (100 ng/ml), anti-FGF2 (2.5 mg/ml), anti-SCF (1 mg/ml), anti-PEDF (2.5 mg/ml), CBOP11 (1.3 mM/ml), BMP4(20 ng/ml), Activin (100 ng/ml)] or their cocktail.

BrdU-treated cells were fixed in 70% ethanol overnight, denatured with 2 N HCl and 0.5% triton in 1× PBS, and neutralized with 1 M boric acid solutions. Cells were washed once with 1% BSA in PBS, incubated in BrdU antibody in blocking solution (1% BSA with 0.5% triton in PBS) for 1 h. Cells were washed and incubated in 10 μg/ml of RNase and 20 μg/ml of propidium iodide (PI) for 30 min at room temperature. FACS analysis was carried out using the BD FACSCalibur.

E18 neurospheres were dissociated into single cells and subjected to Hoechst dye efflux assay, as previously described (Bhattacharya et al., 2003). Briefly, cells (1×10⁶ cells/ml) were incubated in 1 ml of Iscove's modified Dulbecco's medium (IMDM) and 2% FBS overnight. Cells were stained with 3-4 μg/ml of Hoechst 33342 dye for 30 min at 37°C in a shaking water bath. Verapamil (100 μM) and PI controls were included for each experiment. The SP regions were defined based on fluorescence emission in both blue and red wavelengths.

Total RNA was isolated from control and ECCM neurospheres, and used to synthesize biotin-labeled cRNA probe using Gene Chip 30 IVT Express kit (Affymetrix). Fragmented cRNA probes were hybridized to rat genome 430 2.0 gene chip arrays (Affymetrix) at 45°C for 16 h. Arrays were scanned with Affymetrix GCS3000 7G device; images were analyzed with GCOS software. Raw data for each sample were processed separately by robust multiarray analysis using Genepattern analysis software (Reich et al., 2006); log2-transformed values were used to calculate fold changes of specific genes. The accession number for microarray data deposited to NCBI Gene Expression Omnibus Database is GSE5330.

Statistical analysis was performed using an unpaired, two-tailed t-test or one-way analysis of variance (ANOVA) for pairwise and multiple group comparisons, respectively (GraphPad Prism Software). P values less than 0.05 were considered significant. Tukey's method for multiple comparison was used wherever ANOVA showed a significant P value.

The authors thank Dr Mahendra Rao for guidance and consultation, Dr Graham Sharp for constructive criticisms, Dr Yoshiki Sasai for Rx antibody, Lynette Smith in the College of Public Health, UNMC for microarray analysis, Melody Montgomery for editorial help and UNMC microarray core facility.